404-not-founds/CoMa_3B_SFT

CoMa SFT数据集是为论文《Compressing then Matching: An Efficient Pre-training Paradigm for Multimodal Embedding》中的监督微调（SFT）数据而设计的。该论文介绍了一种压缩预训练阶段，作为多模态嵌入模型中对比学习的热身阶段，并展示了如何将MLLM转化为具有竞争力的嵌入模型。使用该数据集训练的模型（CoMa-3B和CoMa-7B）在MMEB基准测试中取得了最先进的结果。

The CoMa SFT dataset is designed for supervised fine-tuning (SFT) data in the paper Compressing then Matching: An Efficient Pre-training Paradigm for Multimodal Embedding. The paper introduces a compressed pre-training phase that serves as a warm-up stage for contrastive learning in multimodal embedding models, and demonstrates how an MLLM can be transformed into a competitive embedding model. The models trained with this dataset (CoMa-3B and CoMa-7B) achieve state-of-the-art results on the MMEB benchmark.